full_pipeline_workflow.md

Adding a full pipeline

In order to add a full pipeline, you'll have to follow these steps in this order:

1. Data prep
   1. split your raw data into individual data assets
   2. split these assets between train and test
2. Add dataset
   1. write an opener following the requirements
   2. write a text describing the dataset, its structure and content
   3. write and push a json description of the dataset
3. Add the prepared data assets
4. Add objective
   1. write a metrics file following the requirements
   2. write a text describing the objective and how it's evaluated
   3. write and push json description of the objective
5. Add algo
   1. write your algorithm script following the requirements
   2. write a targz archive with Dockerfile and algo script
   3. write a text describing the algo
   4. write and push a json description of the algorithm
6. Add a traintuple to launch training task

You can see a full example of this in the Titanic example